The agent-native infrastructure layer

Infrastructure for
the agent internet.

An interoperable layer where AI agents can identify themselves, discover services, and act with delegated authority.

Reference architecture
AGENT-NATIVE INFRASTRUCTUREFIG. 01

01 / The foundation

The trust layer
comes first.

The first planned release gives agent developers a way to identify agents, publish their capabilities, and control what they can access.

Identity

Agent Passport

A portable identity that describes who operates the agent, what it can do, and how to verify it.

Discovery

Registry

A machine-readable directory of agents and services. Find a capability and learn how to connect to it.

Delegation

Permission Gateway

Evaluate each action against owner-defined policies. Scope and revoke access, request approval, and record decisions.

02 / Interactive concept

Follow an
agent request.

A research agent needs a service to summarize a document. See how it proves its identity, finds a service, and requests permission to use it.

See the full ecosystem
Request traceIllustrative run

Agent Passport

Establish who is acting.

The agent presents a credential tied to its owner. The service checks it before accepting the request.

Agent
agt_example_01
Owner
org_example
Credential
Task-scoped
Identity verified
Identify stage selected

03 / Ecosystem roadmap

One layer.
Six product areas.

Identity and the Network establish how agents connect and act. The roadmap extends that foundation to compute, data, spending, and model operations.

Identity

Agent profiles, credentials, and delegated authority.

Network

Agent discovery, messaging, and task delegation.

Cloud

Isolated runtimes, agent hosting, and resumable work.

Data

Persistent memory, files, and portable context.

Commerce

Owner-set budgets, service payments, and receipts.

Intelligence

Model routing, evaluation, and observability.

04 / Architecture

Machine-facing.
Human-governed.

Agents use APIs to discover capabilities, present credentials, and request services. People use a dashboard built on those same APIs to set policy and inspect activity.

  1. Human interfaceDashboard, administration, and activity review

    Operators use the same underlying APIs as agents to manage access, review approvals, and inspect traces.

  2. Agentlithic layerIdentity, discovery, policy, and service coordination

    An agent can be hosted independently and still present its identity, find services, and request narrowly scoped access.

  3. Open protocolsA2A for agent collaboration, MCP for tools, HTTPS and OAuth

    A2A describes how agents communicate and exchange work. MCP connects agents to tools and data. HTTPS and OAuth supply familiar transport and authorization patterns.

  4. Internet and cloudExisting networks, compute, storage, and payment providers

    The planned infrastructure works over the existing internet and integrates with providers for runtime, storage, and payments.